Designing models to predict the weather or the spread of disease can be a great way to increase understanding and accurately predict the future. In the same way, neuroscientists are now beginning to show that our brains may be constantly constructing models of the physical world in order to guide our behaviour and predict what is about the happen. This concept was first developed in Cambridge in the 1940s by Kenneth Craik, who suggested that organisms carry a ‘small-scale model’ of external reality within their brains. In this view an organism is not just physically situated in its environment, but also has an internal model of it, which allows it to deal with external reality in a more effective manner. To construct such a model requires the brain to capture relationships between events in its world. Starting in the 1950s the Cambridge neuroscientist Horace Barlow pioneered the use of concepts from information theory and statistics to understand the brain. He showed how the brain was able to establish whether multiple events occurred spuriously by chance or were related and hence should be incorporated into its model. His statistical approach profoundly influenced the understanding of many areas of neuroscience as well as being a guiding principle for the development of machine learning techniques.

Work in several Cambridge departments continues this approach. In the Zoology Department, insect neurobiologists have elucidated simple neural circuits that enable an animal to eliminate signals that can be predicted from its own actions. This explains why, for example, crickets are not deafened by their own song. In the Department of Physiology, Development & Neuroscience it has been discovered that the brain uses the statistics of previous signals to decide where the eye should look next. In the Department of Engineering the Biological Learning group illustrated how the human brain constructs predictive models of the world and how failure in such prediction can account for delusions of control seen in patients with schizophrenia. All of this research draws on theoretical work in machine learning and statistical inference being carried out in both the Department of Engineering and the Department of Physics.

Viewing the brain from a statistical viewpoint has the potential to unify our understanding of what brains do. Cambridge is exceptionally well placed to bring together the theoretical and experimental approaches necessary to understand brains as statistical machines.